| |
| """Reference-only helper: SenseVoice ASR on AX650 (libax_asr_api) for QA.""" |
| import ctypes, os, sys |
| import numpy as np |
|
|
| _BASE = os.environ.get("ASR_BASE", "/root/huyuan/183_workspace/Speech-Translation-ZipVoice") |
| LIB = os.environ.get("ASR_LIB", _BASE + "/third_party/asr/lib/libax_asr_api.so") |
| MODEL = os.environ.get("ASR_MODEL", _BASE + "/models/asr_official/sensevoice_ax650") |
| sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) |
| from cutetts_ax import audio as A |
|
|
| def main(): |
| lib = ctypes.CDLL(LIB) |
| lib.AX_ASR_Init.argtypes = [ctypes.c_int, ctypes.c_char_p] |
| lib.AX_ASR_Init.restype = ctypes.c_void_p |
| lib.AX_ASR_RunPCM.argtypes = [ctypes.c_void_p, ctypes.POINTER(ctypes.c_float), ctypes.c_int, |
| ctypes.c_int, ctypes.c_char_p, ctypes.POINTER(ctypes.c_char_p)] |
| lib.AX_ASR_RunPCM.restype = ctypes.c_int |
| lib.AX_ASR_Free.argtypes = [ctypes.c_char_p] |
| os.environ.setdefault("AX_SENSEVOICE_SHARE_DIR", MODEL) |
| h = lib.AX_ASR_Init(4, MODEL.encode()) |
| assert h, "init failed" |
| for path in sys.argv[1:]: |
| wav, sr = A.read_wav(path) |
| if sr != 16000: |
| wav = A.resample_to(wav, sr, 16000) |
| arr = np.ascontiguousarray(wav, dtype=np.float32) |
| out = ctypes.c_char_p() |
| rc = lib.AX_ASR_RunPCM(h, arr.ctypes.data_as(ctypes.POINTER(ctypes.c_float)), arr.size, |
| 16000, b"auto", ctypes.byref(out)) |
| txt = out.value.decode("utf-8", "replace") if out.value else "" |
| if out.value: |
| lib.AX_ASR_Free(out) |
| print(f"{os.path.basename(path)}: rc={rc} {txt.strip()}") |
|
|
| if __name__ == "__main__": |
| main() |
|
|
